An on-line biped mini-robot motion learning using neural network and database management

In this research, the back-propagation neural network system and the ACCESS database management are implemented for on-line learning of the ankle joint of the biped robot walking forward. The system acquires the inclining angles of the center of gravity by the inclination sensor and the joint angles by the AI servo motors when the manual teaching is applied for the on-line learning templates of the biped robot walking motions. Regarding the method and process of research, the manual teaching is approached to the motion trajectory of a biped robot, and then all rotating angles of robot joints and the inclining information of the center of gravity are detected via the sensors to be saved. Repeat the original operation route and reconfirm whether the motion trajectory is taught to fall within the center of gravity for regarding as training sample parameters for the motion trajectory. Thus, the motion trajectories of the hip joint angle, the knee joint angle, and the inclining angles of the center of gravity are conducted to approach the ankle joint through the training of the back-propagation neural network system to assure the center of gravity located inside of range of the soles of the feet. The experimental results are shown that the biped min-robot can automatically adjusted the center of body gravity during the walking motion through the back-propagation neural network system.